Knowing where and why species are distributed as they are is foundational to studies of biodiversity and biodiversity monitoring. Increasing access to high resolution remote sensing data, and availability of massive citizen-science observation databases are supporting unprecedented opportunities to map the distribution of organisms at ever finer spatial and temporal resolutions, and at global extent. Currently, most global maps depicting the distribution of biodiversity are generated by overlaying expert-based range maps (i.e., polygons draw on maps, describing where ‘expert(s)’ agree that the species exists). While expert maps are generally accurate over broad spatial scales, they do not perform well predicting species occurrence at fine spatial resolutions, for example, within the expected distribution. By combining expert-based range maps, and point-observation data from collections-based and citizen science projects in an integrated modeling framework, we “update” expert range maps to fine spatial resolution across species’ geographic distributions. We compile range maps, and millions of observation records curated through the map of life project (mol.org) using African sunbirds (family Nectariniidae) as a case-study. In a model selection framework, we used presence-background Poisson point process models with expert maps encoded as offsets, 30m resolution climate and landcover predictors, and estimated with lasso regularization, to model species occurrence for approximately 70 species.
Results/Conclusions
Overall concordance of observation data and expert-based maps (i.e., percentage of observation records found inside versus outside of the expert-based range map) was highly variable across African sunbird species, ranging from approximately 20% of observations occurring within the species’ expert range maps to 99% of points observed within their expert maps. Approximately 12% of species were data deficient in point observations and could not be modeled, and our model selection procedure could not find a model with adequate performance characteristics for approximately 7% of species. Of species not data deficient, we could find adequate-performing models for 93% of species, resulting in occurrence predictions at fine spatial resolution across the entire species’ range. We generated high-resolution maps of species richness and rarity by overlaying high-resolution occurrence information from modeled species, and including expert range maps for species without high resolution models. Expanding this framework to include all species of interest for specific conservation targets can offer opportunities for evidence-based strategic conservation planning under climate or land use change. High-resolution biodiversity products can be used to improve spatial prioritization at fine-scale to efficiently achieve conservation targets.